Tuberculosis has been a major cause of morbidity and mortality globally for hundreds of years and remains so today. In spite of evidence pointing to significant burdens of asymptomatic but infectious (subclinical) disease, there is very little knowledge on the rates and directions of progression. Models reflect this lack of information by simplifying the structure of disease, with very little data informing the direct transitions between states. This thesis seeks to understand the natural history of tuberculosis from first infection to death or recovery with a focus on accurately representing the spectrum of disease through pathology, bacteriology, and symptom presentation in order to better parameterise models. A systematic review of TB research following untreated cohorts found research from the first half of the 1900s describing both progression and regression of disease in terms of bacteriology, radiology, and symptoms. I used this data to construct and parameterise a model of pulmonary TB disease, extending the commonly used “active disease” compartment into three stages of disease: minimal, subclinical, and clinical. I then simulated cohorts through disease to discover that 50% of people with subclinical disease may never present symptomatically, and that, despite finding a median duration of infectious disease of 12 months, after 5 years up to 20% of people starting with infectious disease could still be living with infectious TB. I then used the new model structure to compare the impact different screening tests could provide. With the current limited data on test performance for minimal and subclinical disease, mass x-ray screenings, as used in the first half of the 20th century, are the most impactful without screening every person bacteriologically. A test that could accurately confirm diagnosis of minimal disease could increase the impact of x-ray screening even further. Finally, I parameterised an extension to the TB disease model, including progression from infection to disease. The data for this came from previously unused cohorts in the original systematic review. I found that rates of progression to disease are low in comparison to rates of recovery from infection. The best parameterisations maintained a well-used TB model structure of an intermediate state between infection and disease, along with rapid progression that bypasses this intermediate state. In conclusion, I have created a more complex TB model structure that differentiates between presentations of disease, and based the parameterisation, from infection to death, on real world data. I have used this new structure to evaluate the impact of different screening tests, but there are countless other applications of this model structure for future implementations, including calculating better estimates of the global burden of TB and designing trials.